 RESEARCH AND PRACTICE 
Risk Factors for Femicide in Abusive Relationships:
Results From a Multisite Case Control Study
| Jacquelyn C. Campbell, PhD, RN, Daniel Webster, ScD, MPH, Jane Koziol-McLain, PhD, RN, Carolyn Block, PhD, Doris Campbell, PhD, RN, Mary Ann
Curry, PhD, RN, Faye Gary, PhD, RN, Nancy Glass, PhD, MPH, RN, Judith McFarlane, PhD, RN, Carolyn Sachs, MD, MPH, Phyllis Sharps, PhD, RN,
Yvonne Ulrich, PhD, RN, Susan A. Wilt, DrPH, Jennifer Manganello, PhD, MPH, Xiao Xu, PhD, RN, Janet Schollenberger, MHS, Victoria Frye, MPH,
and Kathryn Laughon, MPH
Femicide, the homicide of women, is the lead-
ing cause of death in the United States among
young African American women aged 15 to
45 years and the seventh leading cause of
premature death among women overall.1
American women are killed by intimate part-
ners (husbands, lovers, ex-husbands, or ex-
lovers) more often than by any other type of
perpetrator.2–4 Intimate partner homicide ac-
counts for approximately 40% to 50% of US
femicides but a relatively small proportion of
male homicides (5.9%).1,5–10 The percentage
of intimate partner homicides involving male
victims decreased between 1976 and 1996,
whereas the percentage of female victims in-
creased, from 54% to 72%.4
The majority (67%–80%) of intimate part-
ner homicides involve physical abuse of the
female by the male before the murder, no
matter which partner is killed.1,2,6,11–13 There-
fore, one of the major ways to decrease inti-
mate partner homicide is to identify and in-
tervene with battered women at risk. The
objective of this study was to specify the risk
factors for intimate partner femicide among
women in violent relationships with the aim
of preventing this form of mortality.
METHODS
An 11-city case–control design was used;
femicide victims were cases (n = 220), and
randomly identified abused women residing
in the same metropolitan area were control
women (n = 343). Co-investigators at each site
collaborated with domestic violence advo-
cacy, law enforcement, and medical examiner
offices in implementing the study. Sampling
quotas for cases and control women in each
city were proportionately calculated so that
the cities with the highest annual femicide
rates included the largest number of cases
and control women.
July 2003, Vol 93, No. 7 | American Journal of Public Health
Objectives. This 11-city study sought to identify risk factors for femicide in abusive
relationships.
Methods. Proxies of 220 intimate partner femicide victims identified from police or
medical examiner records were interviewed, along with 343 abused control women.
Results. Preincident risk factors associated in multivariate analyses with increased
risk of intimate partner femicide included perpetrator’s access to a gun and previous
threat with a weapon, perpetrator’s stepchild in the home, and estrangement, espe-
cially from a controlling partner. Never living together and prior domestic violence ar-
rest were associated with lowered risks. Significant incident factors included the vic-
tim having left for another partner and the perpetrator’s use of a gun. Other significant
bivariate-level risks included stalking, forced sex, and abuse during pregnancy.
Conclusions. There are identifiable risk factors for intimate partner femicides. (Am J
Public Health. 2003;93:1089–1097)
Femicide Cases cide cases abstracted, a knowledgeable proxy
All consecutive femicide police or med- was identified and located. In 82% (307/
ical examiner records from 1994 through 373) of these cases, proxies agreed to partici-
2000 at each site were examined to assess pate. Two exclusion criteria, age (18–50
victim–perpetrator relationships. Cases were years) and no previous abuse by the femicide
eligible if the perpetrator was a current or perpetrator, resulted in the elimination of 87
former intimate partner and the case was additional cases (28.3% of 307 cases), with
designated as “closed” by the police (suicide 59 (19.2% of 307 cases) eliminated solely as
by the perpetrator, arrest, or adjudication, a result of the latter criterion.
depending on the jurisdiction). Records were Researchers and doctoral students experi-
abstracted for data specific to the homicide. enced in working with victims of domestic vi-
At least 2 potential proxy informants, indi- olence conducted telephone or in-person in-
viduals knowledgeable about the victim’s re- terviews in English or Spanish; interviews
lationship with the perpetrator, were identi- were 60 to 90 minutes in duration. Both
fied from the records. The proxy who, in the proxies and abused control women were ex-
investigator’s judgment, was the most knowl- cluded if they could speak neither English
edgeable source was then sent a letter ex- nor Spanish.
plaining the study and including researcher
contact information. If no communication was Abused Control Women
initiated by the proxy, study personnel at- Stratified random-digit dialing (up to 6 at-
tempted telephone or (in the few cases in tempts per number) was used to select
which no telephone contact was possible) per- women aged 18 to 50 years who had been
sonal contact. involved “romantically or sexually” in a rela-
If the first proxy was not knowledgeable tionship at some time in the past 2 years in
about details of the relationship, she or he the same cities in which the femicides oc-
was asked to identify another willing potential curred. A woman was considered “abused” if
proxy informant. When a knowledgeable she had been physically assaulted or threat-
proxy was found, informed consent was ob- ened with a weapon by a current or former
tained. In 373 of the 545 (68%) total femi- intimate partner during the past 2 years; we
Campbell et al. | Peer Reviewed | Research and Practice | 1089

identified episodes of abuse with a modified
version of the Conflict Tactics Scale with
stalking items added.11,14
English- and Spanish-speaking telephone
interviewers employed by an experienced
telephone survey firm completed sensitivity
and safety protocol training.15 A total of 4746
women met the age and relationship criteria
and were read the consent statement. Among
these women, 3637 (76.6%) agreed to partic-
ipate, 356 (9.8%) of whom had been physi-
cally abused or threatened with a weapon by
a current or recent intimate partner. Thirteen
abused control women were excluded from
the analysis because they reported that the
injuries from their most severe incident of
abuse were so severe that they thought they
could have died.
Risk Factor Survey Instrument
The interview included previously tested
instruments, such as the Danger Assess-
ment,16,17 and gathered information on demo-
graphic and relationship characteristics, in-
cluding type, frequency, and severity of
violence, psychological abuse, and harass-
ment; alcohol and drug use; and weapon
availability. The Danger Assessment had
been translated to and validated in Spanish in
earlier research; the remainder of the survey
was translated and back-translated by our
Spanish-speaking interviewers and by project
staff in Houston, Los Angeles, and New York.
A factor analysis of the risk items was used in
constructing scales measuring partners’ con-
trolling and stalking behaviors. Each scale
was internally consistent (α = .83 and .75,
respectively).
Data Analysis
Logistic regression was used to estimate
the independent associations between each
of the hypothesized risk factors and the risk
of intimate partner femicide. Because the im-
portance of certain risk factors may not be
detected when their effects are mediated by
more proximal risk factors, we sequentially
added blocks of conceptually similar explana-
tory variables along a risk factor continuum
ranging from most distal (demographic char-
acteristics of perpetrators and victims) to
most proximal (e.g., weapon used in the femi-
cide or most serious abuse incident). Vari-
ables not significantly associated with femi-
1090 | Research and Practice | Peer Reviewed | Campbell et al.
 RESEARCH AND PRACTICE 
cide risk were dropped from subsequent
models. Model coefficients were exponenti-
ated so that they could be interpreted as ad-
justed odds ratios (ORs).
RESULTS
Demographic, background, and relation-
ship variables that differentiated case women
from control women in bivariate analyses are
presented in Tables 1 and 2. Table 3 displays
findings from the series of logistic regression
models. The strongest sociodemographic risk
factor (model 1) for intimate partner femicide
was the abuser’s lack of employment (ad-
justed OR = 5.09; 95% confidence interval
[CI] = 2.74, 9.45). Instances in which the
abuser had a college education (vs a high
school education) were protective against
femicide (adjusted OR = 0.31; 95% CI = 0.12,
0.80), as were instances in which the abuser
had a college degree and was unemployed
but looking for work. Race/ethnicity of
abusers and victims was not independently
associated with intimate partner femicide risk
after control for other demographic factors.
When additional individual-level risk fac-
tors for homicide were added to the model
(model 2), both abuser’s access to a firearm
(adjusted OR = 7.59; 95% CI = 3.85, 14.99)
and abuser’s use of illicit drugs (adjusted
OR = 4.76; 95% CI = 2.19, 10.34) were
strongly associated with intimate partner
femicide, although the abuser’s excessive use
of alcohol was not. Although the abuser’s ac-
cess to a firearm increased femicide risk, vic-
tims’ risk of being killed by their intimate
partner was lower when they lived apart from
the abuser and had sole access to a firearm
(adjusted OR = 0.22). Neither alcohol abuse
nor drug use by the victim was independently
associated with her risk of being killed.
Relationship variables were added in
model 3. Never having lived with the abusive
partner significantly lowered women’s risk of
femicide (OR = 0.39; 95% CI = 0.16, 0.97).
Having been separated from an abusive part-
ner after living together was associated with a
higher risk of femicide (adjusted OR = 3.64;
95% CI = 1.71, 7.78), as was having ever left
or having asked the partner to leave (adjusted
OR = 3.19; 95% CI = 1.70, 6.02). Having a
child living in the home who was not the abu-
American Journal of Public Health | July 2003, Vol 93, No. 7
sive partner’s biological child more than dou-
bled the risk of femicide (adjusted OR = 2.23;
95% CI = 1.13, 4.39). Addition of the rela-
tionship variables resulted in victims’ sole ac-
cess to a firearm no longer being statistically
significant and substantially reduced the ef-
fects of abuser’s drug use.
Variables related to abusive partners’ con-
trolling behaviors and verbal aggression were
added in model 4. The effects of a highly
controlling abuser were modified by whether
the abuser and victim separated after living
together. The risk of intimate partner femi-
cide was increased 9-fold by the combination
of a highly controlling abuser and the cou-
ple’s separation after living together (adjusted
OR = 8.98; 95% CI = 3.25, 24.83). Femicide
risk was increased to a lesser degree when
the abuser was highly controlling but the cou-
ple had not separated (adjusted OR = 2.90;
95% CI = 1.41, 5.97) and when the couple
had separated after living together but the
abuser was not highly controlling (adjusted
OR = 3.10; 95% CI = 1.20, 8.05).
Threatening behaviors and stalking were
added in model 5. Abusers’ previous threats
with a weapon (adjusted OR = 4.08; 95%
CI = 1.91, 8.72) and threats to kill (adjusted
OR = 2.60; 95% CI = 1.24, 5.42) were associ-
ated with substantially higher risks for femi-
cide. After control for threatening behaviors,
there were no significant independent effects
of abusers’ drug use (OR = 1.64; 95% CI =
0.88, 3.04). The effects of high control with
separation (adjusted OR = 4.07; 95% CI =
1.33, 12.4) and access to guns (adjusted
OR = 5.44; 95% CI = 2.89, 10.22), although
substantially reduced, remained strong.
Stalking and threats to harm children and
other family members were not indepen-
dently associated with intimate partner femi-
cide risk after variables had been entered in
the first models. When variables related to
previous physical abuse were included in
model 6, previous arrest of the abuser for do-
mestic violence was associated with a de-
creased risk of intimate partner femicide (ad-
justed OR = 0.34; 95% CI = 0.16, 0.73). The
association between abusers’ use of forced
sex on victims and increased intimate partner
femicide risks approached statistical signifi-
cance (adjusted OR = 1.87; 95% CI = 0.97,
3.63; P < .07).
  RESEARCH AND PRACTICE 

Incident-level variables were added in

TABLE 1—Sociodemographic Characteristics of Victims and Perpetrators and General Risk model 7. Abuser’s use of a gun in the worst
Factors for Homicide, by Group incident of abuse was associated with a 41-
Victims Perpetrators fold increase in risk of femicide after control
Nonfatal Nonfatal for other risk factors, this effect apparently
Physical Abuse Homicide Physical Abuse Homicide mediating the effects of abuser’s access to a
(n = 343) (n = 220) P (n = 343) (n = 220) P gun, which was no longer significant. How-
ever, previous threats with a weapon contin-
Sociodemographic variables
ued to be associated with increased femicide
Age, y, mean ± SD 30.1 ± 8.6 31.4 ± 7.7 .081 31.2 ± 9.2 34.2 ± 8.7 <.001
risks (OR = 4.41; 95% CI = 1.76, 11.06).
Don’t know/refused/missing 0 0 4 22
When the worst incident of abuse was
Race/ethnicity, No. (%) <.001 <.001
triggered by the victim’s having left the
Black/African American 70 (20.6) 104 (47.3) 83 (24.3) 107 (48.9)
abuser for another partner or by the abuser’s
White 157 (46.3) 53 (24.1) 153 (44.7) 49 (22.4)
jealousy, there was a nearly 5-fold increase
Latino/Hispanic 82 (24.2) 53 (24.1) 80 (23.4) 58 (26.5)
in femicide risk (adjusted OR = 4.91; 95%
Other 30 (8.9) 10 (4.5) 26 (7.6) 5 (2.3)
CI = 2.42, 9.96). When the incident was trig-
Don’t know/refused/missing 4 0 1 1
gered by the victim’s having left the abuser
Education, No. (%) <.001 <.001
for any other reason, femicide risks were
Less than high school 61 (17.9) 71 (33.2) 92 (28.0) 70 (48.9)
also significantly increased (adjusted OR =
High school 73 (21.5) 59 (27.5) 91 (27.7) 47 (32.9)
4.04; 95% CI = 1.80, 9.06). These incident-
Some college/trade school 109 (32.1) 68 (31.8) 58 (17.7) 17 (11.9)
level effects appear to mediate those related
College/trade school 97 (28.5) 16 (7.5) 87 (26.5) 9 (6.3)
to highly controlling abusers and separation
Don’t know/refused/missing 3 6 15 77
after cohabitation.
Employment, No. (%) <.001 <.001
Each of the models included in Table 3
Full-time 179 (52.2) 114 (51.8) 229 (68.2) 84 (39.6)
demonstrated an adequate fit according to
Part-time 70 (20.4) 31 (14.1) 39 (11.6) 20 (9.5) Hosmer–Lemeshow18 goodness-of-fit tests.
Unemployed, seeking job 40 (11.7) 12 (5.5) 25 (7.4) 13 (6.1) Model 6 correctly predicted the case status of
Unemployed, not seeking job 54 (15.7) 63 (28.6) 43 (12.8) 95 (44.8) 73% of the cases and 93% of the control
Don’t know/refused/missing 0 0 7 8 women. Model 7 correctly predicted the case
Income (annual household), $, .005 status of 81% of the cases and 95% of the
No. (%) control women.
Less than 10 000 67 (21.7) 25 (18.8)
10 000–19 999 49 (15.9) 32 (24.1) DISCUSSION
20 000–29 999 43 (13.9) 20 (15.0)
30 000–39 999 41 (13.3) 29 (21.8) Seventy-nine percent (220/279) of the
40 000 or more 109 (35.3) 27 (20.3) femicide victims aged 18 to 50 years and
Don’t know/refused/missing 34 87 70% of the 307 total femicide cases were
General violence/homicide risk variables physically abused before their deaths by the
Threatened/attempted suicide .091 .149 same intimate partner who killed them, in
Yes 33 (9.6) 12 (5.6) 68 (20.1) 45 (25.0) comparison with 10% of the pool of eligible
Don’t know/refused/missing 0 6 4 40 control women. Thus, our first premise, that
Problem alcohol drinker, No. (%) < .001 < .001 physical violence against the victim is the pri-
Yes 27 (7.9) 36 (19.1) 106 (30.9) 105 (52.0) mary risk factor for intimate partner femicide,
Don’t know/refused/missing 0 32 0 18 was upheld. The purpose of this study, how-
Illicit drug use, No. (%) .002 < .001 ever, was to determine the risk factors that,
Yes 49 (14.3) 48 (25.3) 101 (30.4) 123 (65.4) over and above previous intimate partner vio-
Don’t know/refused/missing 1 30 11 32 lence, are associated with femicide within a
Access to a firearm,a No. (%) .996 <.001 sample of battered women. Our analysis
Yes 17 (5.0) 10 (5.0) 82 (23.9) 143(65.0) demonstrated that a combination of the most
Don’t know/refused/missing 2 19 0 0 commonly identified risk factors for homicide,
in conjunction with characteristics specific to
Continued
violent intimate relationships, predicted inti-
mate partner femicide risks.

July 2003, Vol 93, No. 7 | American Journal of Public Health Campbell et al. | Peer Reviewed | Research and Practice | 1091

TABLE 1—Continued
Arrest for violent crime, No. (%)
Yes
Don’t know/refused/missing
Note. The referent time periods for all risk variables were the year previous to the most abusive event for abused control
women and the year previous to the femicide for femicide victims.
a drop out of mandated intervention pro-
For abused women, gun access was defined as a woman’s sole access to a firearm on the basis of her living apart from her
partner and reporting having a gun in the home; gun access for partner was based on reports of his personal ownership of a
firearm or living in a household with a firearm.
The model-building strategy we used al-
lowed for consideration of different levels of
prevention and the degree to which intimate
partner femicides could be prevented by strat-
egies directed at risk factors for homicide in
general. For example, our analysis and those
of others suggest that increasing employment
opportunities, preventing substance abuse,
and restricting abusers’ access to guns can po-
tentially reduce both overall rates of homicide
and rates of intimate partner femicide.
In comparing our femicide perpetrators
with other abusive men, we found that unem-
ployment was the most important demo-
graphic risk factor for acts of intimate partner
femicide. In fact, abuser’s lack of employment
was the only demographic risk factor that sig-
nificantly predicted femicide risks after we
controlled for a comprehensive list of more
proximate risk factors, increasing risks 4-fold
relative to the case of employed abusers
(model 6). Unemployment appears to under-
lie increased risks often attributed to race/
ethnicity, as has been found and reported in
other analyses related to violence.19,20
The present results revealed that traits of
perpetrators thought to be characteristic of vi-
olent criminals in general21 tended to be no
more characteristic of femicide perpetrators
than of other batterers. For instance, in con-
trast to results of previous research compar-
ing abusers and nonabusers,22 our regression
analyses showed that arrests for other crimes
did not differentiate femicide perpetrators
from perpetrators of intimate partner vio-
lence. After controlling for other risk factors,
prior arrest for domestic violence actually de-
creased the risk for femicide, suggesting that
arrest of abusers protects against future inti-
mate partner femicide risks. Perpetrator drug
abuse significantly increased the risk of inti-
1092 | Research and Practice | Peer Reviewed | Campbell et al.
 RESEARCH AND PRACTICE 
<.001
38 (11.5) 43 (21.8)
12 23
mate partner femicide, but only before the ef-
fects of previous threats and abuse were
added. Drug abuse, therefore, was associated
with patterns of intimate partner abuse that
increase femicide risks.
Our iterative model-building strategy also
allowed us to observe whether the effects of
more proximate risk factors mediate the ef-
fects of more distal factors in a manner con-
sistent with theory. For example, the 8-fold in-
crease in intimate partner femicide risk
associated with abusers’ access to firearms at-
tenuated to a 5-fold increase when character-
istics of the abuse were considered, including
previous threats with a weapon on the part of
the abuser. This suggests that abusers who
possess guns tend to inflict the most severe
abuse.
However, consistent with other re-
search,3,23,15,24,25 gun availability still had sub-
stantial independent effects that increased
homicide risks. As expected, these effects
were due to gun-owning abusers’ much
greater likelihood of using a gun in the worst
incident of abuse, in some cases, the actual
femicide. The substantial increase in lethality
associated with using a firearm was consistent
with the findings of other research assessing
weapon lethality. A victim’s access to a gun
could plausibly reduce her risk of being
killed, at least if she does not live with the
abuser. A small percentage (5%) of both case
and control women lived apart from the
abuser and owned a gun, however, and there
was no clear evidence of protective effects.
Previous arrests for domestic violence was
protective against intimate partner femicide
in both of the final models. In most of the
cities where data were collected, there is a
coordinated community response to domes-
tic violence. Under optimal conditions, such
American Journal of Public Health | July 2003, Vol 93, No. 7
responses include adequate and swift adjudi-
cation, close supervision of parole outcomes
through periodic court reviews or specialized
probation programs, ongoing risk manage-
ment for arrested perpetrators and ongoing
safety planning for victims, and close super-
vision involving sanctions for batterers who
grams.26 Under these kinds of conditions,
arrest can indeed be protective against do-
mestic violence escalating to lethality.
Two relationship variables remained signif-
icant throughout the models. Consistent with
earlier research,27,28 instances in which a
child of the victim by a previous partner was
living in the home increased the risk of inti-
mate partner femicide. Situations in which
the victim and abuser had never lived to-
gether were protective, validating safety ad-
vice that battered women have offered to
other battered women in interview studies.29
Women who separated from their abusive
partners after cohabitation experienced in-
creased risk of femicide, particularly when
the abuser was highly controlling. Other stud-
ies have revealed the same risks posed by es-
trangement,30,31 but ours further explicates
the findings by identifying highly controlling
male partners as presenting the most danger
in this situation. At the incident level, we
found that batterers were significantly more
likely to perpetrate homicide if their partner
was leaving them for a different partner.
The bivariate analysis supported earlier ev-
idence that certain characteristics of intimate
partner violence are associated with intimate
partner femicide, including stalking, strangula-
tion, forced sex, abuse during pregnancy, a
pattern of escalating severity and frequency
of physical violence, perpetrator suicidality,
perception of danger on the part of the vic-
tim, and child abuse.15,16,20,32–37 However,
these risk factors, with the exception of forced
sex, were not associated with intimate partner
femicide risk in the multivariate analysis.
Many of these characteristics of abuse are as-
sociated with previous threats with a weapon
and previous threats to kill the victim, factors
that more closely predict intimate partner
femicide risks.
This investigation is one of the few studies
of intimate partner femicide to include a
control population and, to our knowledge,
  RESEARCH AND PRACTICE 

TABLE 2—Relationship Dynamics, Threatening Behavior, and Abuse Characteristics the first to examine the connection between
relationship variables and specific demo-
Abused Control Homicide Victims graphic characteristics of victims and perpe-
Women (n = 343) (n = 220) P
trators. Perhaps the most important limita-
Relationship variables tion of the study is its necessary reliance on
Age difference, y, mean ± SD 1.1 ± 5.7 2.9 ± 6.4 .001 proxy respondents for data regarding hy-
Length of relationship, No. (%) .023 pothesized risk factors for intimate partner
1 month or less 5 (1.5) 0 femicide cases. Because we obtained data
1 month to 1 year 94 (27.5) 44 (20.0) from control women directly, rather than
1 or more years 243 (71.0) 176 (80.0) from a proxy, observed differences between
Don’t know/refused/missing 1 0 case and control women may have been
Relationship partner, No. (%) .005 wholly or partly attributable to differences in
Husband 101 (29.7) 85 (39.0) accuracy of reporting between victims and
Boyfriend 86 (25.3) 65 (29.8) their proxies. To examine this issue, we con-
Ex-husband 36 (10.6) 20 (9.2) ducted a small pilot study comparing re-
Ex-boyfriend 117 (34.4) 48 (22.0) sponses of victims of attempted femicide and
Don’t know/refused/missing 3 2 responses of their proxy respondents and
Separated, No. (%) <.001 found good agreement between summed
Yes 117 (34.9) 101 (55.2) Danger Assessment scores from the 2
Don’t know/refused/missing 8 37 sources of information. Furthermore, there
Cohabitation, No. (%) <.001 was no clear tendency for proxies to under-
Yes 174 (50.7) 81 (45.0) report or overreport victims’ exposure to
In the past year, but not currently 39 (11.4) 68 (37.8) specific risk factors relative to the self-
Previously, but not in the past year 11 (3.2) 11 (6.1) reports of victims themselves.35
Never 118 (34.7) 20 (11.1) It is also possible that some of the women
Don’t know/refused/missing 1 40 who were excluded from this analysis be-
Biological child(ren) of victim and partner living in the cause of no record of previous physical vio-
household, No. (%) .034 lence were in fact being abused, unknown to
Yes 98 (28.6) 73 (37.4)
the proxy. However, we found fairly good
Don’t know/refused/missing 0 25
correspondence with police records of previ-
Biological child(ren) of victim, and not of partner, living
ous domestic violence, and, if anything, we
found more knowledge of previous physical
in the household, No. (%) <.001
abuse among proxies than among police. A
Yes 60 (17.5) 82 (38.7)
related limitation is the relatively large pro-
Don’t know/refused/missing 0 8
portion of “don’t know” responses from prox-
Relationship abuse dynamics
ies regarding certain hypothesized risk fac-
Partner controlling behaviors (score > 3), No. (%) <.001
tors of a more personal nature (e.g., forced
Yes 84 (24.5) 145 (65.9)
sex). Our decision to treat these “don’t know”
Partner called victim names to put her down, No. (%) <.001
responses as representing absence of the “ex-
Yes 164 (47.8) 151 (77.8)
posure” produced conservative biases in our
Don’t know/refused/missing 0 26
estimates of relationships with intimate part-
General violence/homicide risk variables
ner femicide risks. Therefore, we may have
Partner violent outside home, No. (%) <.001
inappropriately failed to reject the null hy-
Yes 116 (35.5) 102 (55.7)
pothesis in the case of some of these vari-
Don’t know/refused/missing 16 37
ables with large amounts of missing data and
Partner threatened to kill woman, No. (%) <.001
near-significant associations with intimate
Yes 50 (14.6) 142 (73.6)
partner femicide risk.
Don’t know/refused/missing 1 27
Another limitation was that we excluded
Partner threatened to kill family, No. (%) <.001
women who did not reside in large urban
Yes 26 (7.6) 72 (33.8)
areas (other than Wichita, Kan) and control
Don’t know/refused/missing 0 7
group women who did not have telephones.
Continued We also failed to keep records of exactly
which proxy interviews (estimated to be less

July 2003, Vol 93, No. 7 | American Journal of Public Health Campbell et al. | Peer Reviewed | Research and Practice | 1093
  RESEARCH AND PRACTICE 

TABLE 2—Continued than 10% of the total) were conducted in

person rather than by telephone, and thus
Partner threatened woman with a weapon, No. (%) <.001 we cannot evaluate the effects of this source
Yes 16 (4.7) 110 (55.3) of bias. Finally, we have no way to compare
Don’t know/refused/missing 0 21 the control women who participated with
Partner threatened to harm children, No. (%) <.001 those who did not, and women living in the
Yes 4 (1.2) 36 (18.5) most dangerous situations may have been
Don’t know/refused/missing 7 25 less likely to participate as control women. If
Stalking behavior (score > 3), No. (%) <.001 so, true exposure to the risk factors of inter-
Yes 21 (6.1) 47 (21.4) est among women involved in abusive inti-
Don’t know/refused/missing 0 0 mate relationships may be greater than our
Characteristics of physical violence control data suggest, thus inflating our esti-
Increase in frequency, No. (%) <.001 mates of increased risks associated with
Yes 88 (25.7) 109 (59.9) these exposures.
Don’t know/refused/missing 5 38
Increase in severity, No. (%) <.001 CONCLUSIONS
Yes 70 (20.4) 105 (64.4)
Don’t know/refused/missing 5 57 In light of our findings, it is important to
Partner tried to choke (strangle) woman, No. (%) <.001 consider the role medical professionals might
Yes 34 (9.9) 84 (56.4) play in identifying women at high risk of inti-
Don’t know/refused/missing 1 71 mate partner femicide. The variables that re-
Forced sex, No. (%) <.001 mained significant in model 6 are those most
Yes 51 (14.9) 84 (57.1) important for identifying abused women at
Don’t know/refused/missing 1 73 risk for femicide in the health care system
Abused during pregnancy (ever), No. (%) <.001
and elsewhere, whereas those that were sig-
nificant in model 7 are particularly important
Yes 24 (7.0) 49 (25.8)
in prevention of the lethal incident itself.
No or never been pregnant 319 (93.0) 141 (74.2)
When women are identified as abused in
Don’t know/refused/missing 0 30
medical settings, it is important to assess per-
Partner arrest previously for domestic violence, No. (%) .003
petrators’ access to guns and to warn women
Yes 46 (13.9) 50 (25.6%)
of the risk guns present. This is especially
Don’t know/refused/missing 12 25
true in the case of women who have been
Incident-level variables
threatened with a gun or another weapon
Gun used, No. (%) <.001
and in conditions of estrangement. Under fed-
Yes 3 (0.9) 84 (38.2)
eral law, individuals who have been con-
Partner used alcohol or drugs, No. (%) <.001
victed of domestic violence or who are sub-
Yes 123 (34.6) 133 (60.5)
ject to a restraining order are barred from
Victim used alcohol or drugs, No. (%) <.001
owning firearms. Judges issuing orders of pro-
Yes 44 (12.4) 53 (24.1)
tection in cases of intimate partner violence
Order of protection, No. (%) <.001
should consider the heightened risk of lethal
Yes 16 (4.7) 54 (24.5)
violence associated with abusers’ access to
Trigger: jealousy, No. (%) <.001
firearms.
Yes 52 (17.1) 85 (38.6)
Often, battered women like the idea of a
No or don’t know 291 (82.9) 135 (61.4)
health care professional notifying the police
Trigger: woman leaving, No. (%) <.001
for them; however, with the exception of Cal-
Yes 32 (10.5) 72 (32.7) ifornia, states do not require health care pro-
No or don’t know 311 (89.5) 148 (67.3) fessionals to report to the criminal justice sys-
Trigger: woman has new relationship, No. (%) <.001 tem unless there is evidence of a felony
Yes 7 (2.0) 26 (11.8) assault or an injury from an assault.38–40 In
No or don’t know 336 (98.0) 194 (88.2) states other than California, the professional
Note. Unless otherwise noted, the referent time periods for risk variables were the year previous to the most abusive event for can offer to call the police, but the woman
abused control women and the year previous to the femicide for femicide victims. should have the final say, in that she can
best assess any increased danger that might

1094 | Research and Practice | Peer Reviewed | Campbell et al. American Journal of Public Health | July 2003, Vol 93, No. 7
  RESEARCH AND PRACTICE 

TABLE 3—Hypothesized Risk Factors for Intimate Partner Femicide Among Women Involved result from the police being notified. An ex-
in a Physically Abusive Intimate Relationship Within the Past 2 Years: Adjusted Odds Ratios cellent resource for referral, shelter, and in-
formation is the National Domestic Violence
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7
Hotline (1-800-799-SAFE).
Abuser age 1.10*** 1.08*** NS If a woman confides that she is planning to
Abuser race/ethnicity NS leave her abuser, it is critical to warn her not
Abuser education (reference group: to confront him personally with her decision.
high school graduates) Instead, she needs to leave when he is not
Less than high school 1.40 NS present and leave a note or call him later. It is
Some college 0.72 NS also clear that extremely controlling abusers
College 0.31* NS are particularly dangerous under conditions
Abuser job status (reference group: of estrangement. A question such as “Does
employed full time) your partner try to control all of your daily
Employed part time 1.61 NS NS NS NS NS NS activities?” (from the Danger Assessment15)
Unemployed, seeking job 1.34 NS NS NS NS NS NS can quickly assess this extreme need for con-
Unemployed, not seeking job 5.09*** 6.27*** 4.00*** 3.24*** 4.28*** 4.42*** 4.35* trol. Health care professionals can also expe-
Victim age NS ditiously assess whether the perpetrator is un-
Victim race/ethnicity NS employed, whether stepchildren are present
Victim education (reference group: in the home, and whether the perpetrator has
high school graduates) threatened to kill the victim. Under these con-
Less than high school 1.61 NS NS NS ditions of extreme danger, it is incumbent on
Some college 0.87 NS NS NS health care professionals to be extremely as-
College 0.31** 0.15* 0.28* NS sertive with abused women about their risk of
Victim job status (reference group: homicide and their need for shelter.41
employed full time)
Employed part time 0.95 NS NS
Unemployed, seeking job 0.13*** 0.25* NS About the Authors
Unemployed, not seeking job 0.99 NS NS Jacquelyn C. Campbell, Phyllis Sharps, and Kathryn
Laughon are with the School of Nursing, Johns Hopkins
General risk factors for homicide University, Baltimore, Md. Daniel Webster, Jennifer
Abuser problem drinker NS Manganello, and Janet Schollenberger are with the
Abuser used illicit drugs 4.76*** 2.19* 1.88* NS NS Bloomberg School of Public Health, Johns Hopkins Uni-
versity. Jane Koziol-McLain is with the School of Nursing,
Abuser mental health NS
Auckland University of Technology, Auckland, New
Abuser threatened suicide NS Zealand. Carolyn Rebecca Block is with the Illinois Crim-
Abuser hurt pet NS inal Justice Information Authority, Chicago. Doris Camp-
bell is with the College of Medicine, University of South
Abuser access to gun 7.59*** 9.21*** 8.28*** 5.44*** 5.38*** NS
Florida, Tampa. Mary Ann Curry and Nancy Glass are
Abuser arrest for violent crime NS with the School of Nursing, Oregon Health Sciences Uni-
Victim problem drinker NS versity, Portland. Faye Gary is with the College of Nurs-
ing, University of Florida, Gainesville. Judith McFarlane
Victim used illicit drugs NS
is with the School of Nursing, Texas Women’s University,
Victim sole access to gun 0.22* NS NS NS NS NS Houston. Carolyn Sachs is with the School of Medicine,
Relationship variables University of California Los Angeles. Yvonne Ulrich is
with the School of Nursing, University of Washington,
Married NS
Seattle. Susan A. Wilt is with the New York City Depart-
Divorced NS ment of Health. Xiao Xu is with Covance Inc, Washing-
Time in relationship NS ton, DC. Victoria A. Frye is with St. Luke’s Medical Cen-
ter, New York City.
Cohabitation (reference: living
Requests for reprints should be sent to Jacquelyn C.
together during entire past Campbell, PhD, RN, Johns Hopkins University, School of
year) Nursing, 525 N Wolfe St, #436, Baltimore, MD 21205-
Living together less than 1 year NS 2110 (e-mail: jcampbell@son.jhmi.edu).
This article was accepted September 23, 2002.
Previously lived together, 3.64**
separated at time of
Contributors
incident J. C. Campbell designed the study and wrote most of
Never lived together 0.39** 0.30** 0.36* 0.34** 0.31** the introductory and Discussion sections. D. Webster
analyzed the data, wrote most of the Results section,
Continued and contributed to the Methods and Discussion sec-
tions. J. Koziol-McLain wrote the Methods section, con-

July 2003, Vol 93, No. 7 | American Journal of Public Health Campbell et al. | Peer Reviewed | Research and Practice | 1095

TABLE 3—Continued
Victim left or asked abuser to leave
Victim–abuser had biological child
Victim had child by a previous
partner in home
Abuser–victim age difference
Abuser control of victim, verbal
aggression
Calls names
Not high control and separated
after living together
High control and not separated
after living together
High control and separated after
living together
Abuser threats and stalking
Threatened to harm children
Threatened to harm family
Threatened victim with weapon
Threatened to kill victim
Stalking
Physical abuse before worst incident
Abuse increasing in frequency
and severity
Choked (strangled)
Forced sex
Abused when pregnant
Previous arrest for domestic
violence
Incident-level risk factors
Abuser used alcohol or drugs
Victim used alcohol or drugs
Abuser used gun
Trigger: jealousy/victim left for
other relationship
Trigger: victim left abuser for
other reasons
Note. NS = nonsignificant.
*P < .05; **P < .01; ***P < .001.
tributed to the Results section, and prepared the tables.
J. Manganello contributed to the data analysis and Re-
sults sections. All other authors collected data, con-
tributed to the introductory and Discussion sections,
and reviewed the article.
Acknowledgments
This research was supported by joint funding from the
National Institute on Alcohol Abuse and Alcoholism,
the National Institute on Drug Abuse, the National In-
stitute of Mental Health, the National Institutes on
Aging, the Centers for Disease Control and Prevention,
and the National Institute of Justice (grant R01 # DA/
AA11156).
1096 | Research and Practice | Peer Reviewed | Campbell et al.
 RESEARCH AND PRACTICE 
References
1. Greenfield LA, Rand MR, Craven D, et al. Vio-
3.20** 2.40** NS lence by Intimates: Analysis of Data on Crimes by Current
or Former Spouses, Boyfriends, and Girlfriends. Washing-
NS
ton, DC: US Dept of Justice; 1998.
2.23** 1.70 1.94* 2.44** 2.35*
2. Mercy JA, Saltzman LE. Fatal violence among
spouses in the United States: 1976–85. Am J Public
NS Health. 1989;79:595–599.
3. Bailey JE, Kellermann AL, Somes GW, Banton JG,
Rivara FP, Rushforth NP. Risk factors for violent death
of women in the home. Arch Intern Med. 1997;157:
NS
777–782.
3.10* 3.36* 3.64* 3.10*
4. Bachman R, Saltzman LE. Violence Against
Women: Estimates From the Redesigned Survey. Washing-
2.90** 2.09* 2.08* 2.40* ton, DC: Bureau of Justice Statistics; 1995.
5. Browne A, Williams KR, Dutton DC. Homicide
8.98*** 4.07* 5.52** 3.43* between intimate partners. In: Smith MD, Zah M, eds.
Homicide: A Sourcebook of Social Research. Thousand
Oaks:Sage,1998:149–164.
6. Langford L, Isaac NE, Kabat S. Homicides related
NS to intimate partner violence in Massachusetts. Homicide
NS Stud. 1998;2:353–377.
4.08*** 3.38*** 4.41* 7. Moracco KE, Runyan CW, Butts J. Femicide in
2.60** 3.22** NS North Carolina. Homicide Stud. 1998;2:422–446.
NS 8. Frye V, Wilt S, Schomburg D. Female homicide in
New York City, 1990–1997. Available at: http://www.
nyc.gov/html/doh/pdf/ip/female97.pdf. Accessed Au-
NS gust 18, 2002.
9. National Institute of Justice. A Study of Homicide
NS in Eight US Cities: An NIJ Intramural Research Project.
Washington, DC: US Dept of Justice; 1997.
1.87 NS
NS 10. Wilt SA, Illman SM, Brodyfield M. Female Homi-
cide Victims in New York City. New York, NY: New York
0.34** 0.31* City Dept of Health; 1995.
11. Campbell JC. “If I can’t have you, no one can”:
power and control in homicide of female partners. In:
NS Radford J, Russell DEH, eds. Femicide: The Politics of
Woman Killing. New York, NY: Twayne; 1992:99–113.
NS
41.38** 12. McFarlane J, Campbell JC, Wilt S, Sachs C, Ulrich
Y, Xu X. Stalking and intimate partner femicide. Homi-
4.91*** cide Stud. 1999;3:300–316.
13. Pataki G. Intimate Partner Homicides in New York
4.04*** State. Albany, NY: New York State Governor’s Office;
1997.
14. Straus MA, Gelles RJ. Physical Violence in Ameri-
can Families: Risk Factors and Adaptations to Family Vi-
olence in 8,145 Families. New Brunswick, NJ: Transac-
tion Publishers; 1990.
15. Johnson H, Sacco VF. Researching violence
We would like to thank our advocacy, criminal jus- against women: Statistics Canada’s national survey. Can
tice, and medical examiner collaborators at each of the J Criminology. 1995;37:281–304.
sites, along with the women and family members who
told their stories. We also thank Arthur Kellerman, 16. Campbell JC. Prediction of homicide of and by
MD, for his wise consultation and original ideas. Fi- battered women. In: Campbell JC, ed. Assessing the Risk
nally, we thank the staff of the Data Stat Survey Re- of Dangerousness: Potential for Further Violence of Sexual
search Firm and Jo Ellen Stinchcomb, Nadiyah John- Offenders, Batterers, and Child Abusers. Newbury Park,
son, and the many other assistants and students for all Calif: Sage Publications; 1995:93–113.
of their work. 17. Campbell JC, Sharps P, Glass NE. Risk assessment
for intimate partner violence. In: Pinard GF, Pagani L,
eds. Clinical Assessment of Dangerousness: Empirical
Human Participant Protection Contributions. New York, NY: Cambridge University
Institutional review board approval was obtained from
Press; 2000:136–157.
each study site. Informed consent was obtained by tele-
phone from all participants who were interviewed. 18. Hosmer DW, Lemeshow S. A goodness-of-fit test
American Journal of Public Health | July 2003, Vol 93, No. 7
  RESEARCH AND PRACTICE 

for the multiple logistic regression model. Commun Stat. vivors’ predictions. J Interpersonal Violence. 2000;15:
1980;A10:1043–1069. 75–90.
19. Hawkins DF. Inequality, culture, and interpersonal 37. Saunders DG, Browne A. Intimate partner homi-
violence. Health Aff (Millwood). 1993;12:80–95. cide. In: Ammerman RT, Hersen M, eds. Case Studies
in Family Violence. New York, NY: Kluwer Academic
20. Stets JE. Job autonomy and control over one’s
Publishers; 2000:415–449.
spouse: a compensatory process. J Health Soc Behav.
1995;35:244–258. 38. Chalk R, King P. Violence in families: assessing
prevention and treatment programs. In: Chalk R, King
21. Fagan J, Stewart DE, Hansen K. Violent men or
PA, eds. Health Care Interventions. Washington, DC: Na-
violent husbands? Background factors and situational
tional Academy Press; 1998.
correlates. In: Gelles RJ, Hotaling G, Straus MA, Finkel-
hor D, eds. The Dark Side of Families. Beverly Hills, 39. Gielen AC, O’Campo P, Campbell J, et al. Women’s
Calif: Sage Publications; 1983:49–68. opinions about domestic violence screening and manda-
tory reporting. Am J Prev Med. 2000;19:279–285.
22. Weiner NA, Zahn MA, Sagi RJ. Violence: Patterns,
Causes, Public Policy. New York, NY: Harcourt Brace Jo- 40. Rodriguez MA, McLoughlin E, Nah G, Campbell War and Public
vanovich; 1990.
23. Browne A, Williams KR, Dutton DC. Homicide
JC. Mandatory reporting of domestic violence injuries
to the police: what do emergency department patients Health
think? JAMA. 2001;286:580–583.
between intimate partners. In: Smith MD, Zahn M, eds. by Barry S. Levy and
Homicide: A Sourcebook of Social Research. Thousand 41. Wadman MC, Muelleman RL. Domestic violence
Oaks, Calif: Sage Publications; 1998:149–164. homicides: ED use before victimization. Am J Emerg Victor W. Sidel
Med. 1999;17:689–691.
24. Arbuckle J, Olson L, Howard M, Brillman J, Anctil
C, Sklar D. Safe at home? Domestic violence and other Updated edition with all-new
homicides among women in New Mexico. Ann Emerg
Med. 1996;27:210–215.
epilogue
25. Kellerman AL, Rivara FP, Rushforth NB. Gun
ownership as a risk factor for homicide in the home.
N Engl J Med. 1993;329:1084–1091.
I n this softcover edition, contributors
demonstrate the devastating effects of
war. They discuss nuclear weapons, bio-
26. Gondolf EW. Batterer Intervention Systems: Issues, logical and chemical weapons, conven-
Outcomes, and Recommendations. Thousand Oaks, Calif: tional arms and services, the United
Sage Publications; 2002. Nations, and the enormous costs involved
in depriving warring nations from focus-
27. Daly M, Wiseman KA, Wilson M. Women and
ing on the health and welfare of their citi-
children sired by previous partners incur excess risk of
zens.
uxorcide. Homicide Stud. 1997;1:61–71.
This book should be on the reading list
28. Brewer VE, Paulsen DJ. A comparison of US and of not only health professionals but of all
Canadian findings on uxorcide risk for women with those who are interested in international
children sired by previous partners. Homicide Stud. studies, diplomacy or the military.
1999;3:317–332.
29. Campbell JC, Miller P, Cardwell MM, Belknap RA. ISBN 0-87553-023-0
Relationship status of battered women over time. J Fam 2000 ❚ 417 pages ❚ softcover
Violence. 1994;9:99–111. $17.00 APHA Members
$23.50 Nonmembers
30. Wilson M, Daly M. Spousal homicide risk and es- plus shipping and handling
trangement. Violence Vict. 1993;8:3–15.
31. Dawson R, Gartner R. Differences in the charac-
American Public Health Association
teristics of intimate femicides: the role of relationship Publication Sales
state and relationship status. Homicide Stud. 1998;2:
Web: www.apha.org
E-mail: APHA@TASCO1.com
378–399. Tel: (301) 893-1894
32. Campbell JC, Soeken K, McFarlane J, Parker B. FAX: (301) 843-0159 WR01J7
Risk factors for femicide among pregnant and nonpreg-
nant battered women. In: Campbell JC, ed. Empowering
Survivors of Abuse: Health Care for Battered Women and
Their Children. Thousand Oaks, Calif: Sage Publica-
tions; 1998:90–97.
33. Campbell JC, Soeken K. Forced sex and intimate
partner violence: effects on women’s health. Violence
Women. 1999;5:1017–1035.
34. McFarlane J, Soeken K, Campbell JC, Parker B,
Reel S, Silva C. Severity of abuse to pregnant women
and associated gun access of the perpetrator. Public
Health Nurs. In press.
35. Websdale N. Understanding Domestic Homicide.
Boston, Mass: Northeastern University Press; 1999.
36. Weisz A, Tolman R, Saunders DG. Assessing the
risk of severe domestic violence: the importance of sur-

July 2003, Vol 93, No. 7 | American Journal of Public Health Campbell et al. | Peer Reviewed | Research and Practice | 1097